AI-enabled climate-smart early yield forecasting model for row crops
An AI-driven deep learning yield forecasting system for row crops that identifies yield 'upper and lower limits' using canopy light interception, spectral biotic stress detection, and long-term climate factors, enabling early-stage yield prediction for crops such as cotton, corn, and soybean.
Overview
This solution applies artificial intelligence and deep learning to the challenge of early-stage crop yield forecasting for row crops. The approach is built on the insight that crop yield operates between two limits: an upper limit set by the genetic potential of the cultivar (determined largely by canopy light interception, or photosynthetically active radiation) and a lower limit set by biotic and abiotic stressors that reduce productivity. By identifying both limits, the model can isolate the key determinants of yield and forecast outcomes earlier in the growing season. The system is being adapted for row crops including cotton, corn, and soybean, with potential value for growers, agronomists, seed companies, insurers, and agricultural decision-support providers seeking data-driven yield predictions.
Technical specifications
- Upper limit identification: Quantifies the relationship between photosynthetically active radiation (PAR) and yield, with prior validation showing an R² of up to 0.84 in almond cultivars, providing a baseline for maximum potential yield.
- Lower limit identification: Uses spectral signatures to detect biotic stressors and incorporates long-term climate data to account for abiotic stressors that reduce yield below the upper limit.
- Deep learning architecture: Modified cutting-edge models including ResNet and DeepLab v3 have achieved up to 97% accuracy in object prediction from images, forming the foundation for feature engineering and yield forecasting.
- Temporal modeling: A modified Long Short-Term Memory (LSTM) network integrates accumulated indicators such as annual precipitation alongside internal and external influencing factors.
- Field validation infrastructure: Access to the on-campus Mississippi State University farm enables in-field data collection and model training under real growing conditions.
Technology readiness level
The underlying components have been validated in prior research, with demonstrated correlations between light interception and yield (R² up to 0.84) and high accuracy (up to 97%) in modified deep learning image models. The next stage of validation targets a main row crop in Mississippi over a one-year study period, collaborating with a plant physiologist and seeking a student researcher with a computer science background and agricultural interests. The system is at an early-to-mid technology readiness stage, with proven component-level performance and active development toward a comprehensive integrated forecasting model for row crop production.
About Mississippi State University
Mississippi State University is a comprehensive public land‑grant research university based in Starkville with a large residential campus and a statewide footprint. Industry connects through an adjacent research and technology park that co‑locates corporate tenants with university labs, prototyping spaces, and co‑op and internship pipelines. As the state’s land‑grant, MSU runs an extensive extension network and experiment stations that offer field‑scale testbeds and rapid access to partners across Mississippi. Research is supported by competitive funding from agencies such as NSF, USDA, DoD, and DOE. A dedicated technology transfer office streamlines IP, licensing, and startup formation, with corporate engagement staff shaping sponsored research.
Inquire about this partnering listing
Looking for
Industry applications
delivered to your inbox.